AI progress in the 2021 report
Evidence you can use
AI progress in the 2021 report
Historical snapshot: October 2021. Dates and populations are specified per row.
| Measure | Reported value | Definition and source |
|---|---|---|
| CLIP training pairs | 400M | Text-image pairs used for pretraining.2021 report, slide 38 (PDF page 38) |
| Best model truthfulness | 58% | TruthfulQA comparison reported in the 2021 deck.2021 report, slide 44 (PDF page 44) |
| Human baseline truthfulness | 94% | Human comparison on the same benchmark.2021 report, slide 44 (PDF page 44) |
Training size is not a quality score. TruthfulQA was designed to expose particular failure modes; its rates should not be treated as universal accuracy measures.
Sources and dates
Historical snapshot published October 12, 2021. This web edition was prepared on 2026-10-11 from the online deck and original launch posts. Findings and forecasts retain their original time frame.
- 2021 report, slide 11 (PDF page 11)Original 2021 report. Printed slide numbers match PDF pages.
- 2021 report, slide 29 (PDF page 29)Original 2021 report. Printed slide numbers match PDF pages.
- 2021 report, slide 38 (PDF page 38)Original 2021 report. Printed slide numbers match PDF pages.
- 2021 report, slide 44 (PDF page 44)Original 2021 report. Printed slide numbers match PDF pages.
- State of AI Report 2021: online slides
- Air Street Press launch essayOctober 12, 2021.
- Welcome to State of AI Report 2021Original website launch post, October 12, 2021.
Cite this page
Benaich, Nathan, and Ian Hogarth. “AI progress in the 2021 report.” State of AI Report 2021. Historical report snapshot; web edition prepared 2026-10-11.